Artificial intelligence (AI) models to improve image processing related to pre and post item deliveries

Abstract

Techniques for improving image processing related to item deliveries are described. In an example, a computer system receives an image showing a drop-off of an item, the item associated with a delivery to a delivery location. The computer system inputs the image to a first artificial intelligence (AI) model. The computer system receives first data comprising an indication of whether the drop-off is correct from the first AI model. The computer system causes a presentation of the indication at a device associated with the delivery of the item to the delivery location.

What is claimed is: 
     
       1. One or more non-transitory computer-readable storage media storing instructions that, upon execution on a computer system, cause the computer system to perform operations comprising:
 inputting, to a first artificial intelligence (AI) model, a set of images showing item drop-offs at a delivery location, the first AI model trained for object detection based at least in part on training images showing historical item drop-offs at a plurality of delivery locations; 
 receiving, from the first AI model, first data that identifies objects at the delivery location, relative positioning of the objects, and a correct drop-off location, the first data comprising a label that identifies a first object, a first indication of a location of the first object relative to a second object at the delivery location, and a second indication of the correct drop-off location; and 
 causing, by at least sending the first data to a device associated with a delivery of an item to the delivery location, the device to present the label, the first indication, and the second indication at a user interface of the device. 
 
     
     
       2. The one or more non-transitory computer-readable storage media of  claim 1 , wherein the one or more non-transitory computer-readable storage media store further instructions that, upon execution on the computer system, cause the computer system to perform additional operations comprising:
 inputting, to the first AI model, environmental data associated with the delivery location; 
 receiving, from the first AI model, second data that comprises a safety warning associated with the correct drop-off location; and 
 causing, by at least sending the second data to the device, the device to present the safety warning at the user interface in conjunction with presenting the label, the first indication, and the second indication. 
 
     
     
       3. The one or more non-transitory computer-readable storage media of  claim 1 , wherein the one or more non-transitory computer-readable storage media store further instructions that, upon execution on the computer system, cause the computer system to perform additional operations comprising:
 inputting, to a second AI model, the first data and environmental data associated with the delivery location, the second AI model trained for safety warning recommendations based at least in part on training data that indicates detected objects at the plurality of delivery locations and environmental conditions associated with the plurality of delivery locations; 
 receiving, from the second AI model, second data that comprises a safety warning associated with the correct drop-off location; and 
 sending the second data to the device. 
 
     
     
       4. The one or more non-transitory computer-readable storage media of  claim 1 , wherein the one or more non-transitory computer-readable storage media store further instructions that, upon execution on the computer system, cause the computer system to perform additional operations comprising:
 receiving, from the device, an image showing a drop-off of the item; 
 inputting, to a multi-class classifier, the image based at least in part on second data indicating an incorrect item drop-off, wherein the multi-class classifier is trained based at least in part on (i) the set of images and (ii) additional images showing at least one of: drop-offs at one or more delivery locations within a distance range of the delivery location or other drop-offs at random delivery locations; 
 receiving, from the multi-class classifier, third data indicating at least one of: a second delivery location that corresponds to the incorrect item drop-off or an incorrect delivery at an unknown location; and 
 sending, to at least one of the device or another device, the third data, the third data presented on a display of the at least one of the device or the other device. 
 
     
     
       5. A method implemented on a computer system, the method comprising:
 inputting, to a first artificial intelligence (AI) model, a set of images showing item drop-offs at a delivery location, the first AI model trained for object detection based at least in part on training images; 
 receiving, from the first AI model, first data that identifies objects at the delivery location, relative positioning of the objects, and a correct drop-off location, the first data comprising a label that identifies a first object, a first indication of a location of the first object relative to a second object at the delivery location, and a second indication of the correct drop-off location; and 
 causing, by at least providing the first data to a device associated with a delivery of an item to the delivery location, the device to present the label, the first indication, and the second indication at a user interface of the device. 
 
     
     
       6. The method of  claim 5 , further comprising:
 receiving, from the device, an image showing a drop-off of the item; 
 inputting the image to a second AI model, the second AI model trained for item drop-off classification at the delivery location based at least in part on the set of images; 
 receiving, from the second AI model, second data indicating whether the drop-off is correct; and 
 sending, to at least one of the device or another device, the second data, the second data presented on a display of the at least one of the device or the other device. 
 
     
     
       7. The method of  claim 5 , further comprising:
 generating a labeled image from an image of the set of images, the labeled image showing the objects and comprising a first bounding box around the first object and the label associated with the first bounding box, wherein the first indication comprises the first bounding box; and 
 sending the labeled image to the device, wherein the labeled image is presented at the user interface. 
 
     
     
       8. The method of  claim 7 , wherein the labeled image further comprises a second bounding box around the correct drop-off location. 
     
     
       9. The method of  claim 5 , further comprising:
 determining environmental data associated with the delivery location; 
 generating, based at least in part on the environmental data and the first data, second data that comprises a safety warning associated with the delivery of the item; and 
 sending the second data to the device. 
 
     
     
       10. The method of  claim 9 , wherein generating the second data comprises at least one of:
 inputting the environmental data to the first AI model; or 
 inputting the environmental data and the first data to a second AI model trained for safety warning recommendations. 
 
     
     
       11. The method of  claim 5 , further comprising:
 determining environmental data associated with the delivery location; and 
 determining, based at least in part on the set of images and the environmental data, the correct drop-off location. 
 
     
     
       12. The method of  claim 11 , wherein determining the correct drop-off location comprises inputting the environmental data to the first AI model. 
     
     
       13. The method of  claim 5 , wherein determining the correct drop-off location further comprises inputting, to the first AI model, item data identifying one or more attributes of the item. 
     
     
       14. A computer system comprising:
 one or more processors; and 
 one or more memories storing computer-readable instructions that, upon execution by the one or more processors, configure the computer system to:
 input, to a first artificial intelligence (AI) model, a set of images showing item drop-offs at a delivery location, the first AI model trained for object detection based at least in part on training images showing historical item drop-offs at a plurality of delivery locations; 
 receive, from the first AI model, first data that identifies objects at the delivery location, relative positioning of the objects, and a correct drop-off location, the first data comprising a label that identifies a first object, a first indication of a location of the first object relative to a second object at the delivery location, and a second indication of the correct drop-off location; and 
 cause, by at least sending the first data to a device associated with a delivery of an item to the delivery location, the device to present the label, the first indication, and the second indication at a user interface of the device. 
 
 
     
     
       15. The computer system of  claim 14 , wherein the one or more memories store further computer-readable instructions that, upon execution by the one or more processors, additionally configure the computer system to:
 receive, from the device, an image showing a drop-off of the item; 
 input the image to a second AI model, the second AI model trained for item drop-off classification based at least in part on the set of images; 
 receive, from the second AI model, second data comprising a third indication of whether the drop-off is correct; and 
 cause a presentation of the third indication at the device. 
 
     
     
       16. The computer system of  claim 15 , wherein the first AI model comprises a machine learning (ML) model that is trained for feature extraction based at least in part on the training images, and wherein the second AI model comprises a classifier that is trained for drop-off classifications. 
     
     
       17. The computer system of  claim 16 , wherein inputting the image comprises:
 inputting the image to the ML model; 
 receiving features extracted by the ML model based at least in part on the image; 
 inputting the features to the classifier; and 
 receiving the third indication from the classifier based at least in part on the features. 
 
     
     
       18. The computer system of  claim 15 , wherein the second AI model comprises a multi-class classifier that is trained for delivery location prediction based at least in part on (i) the set of images and (ii) additional images showing other drop-offs at one or more other delivery locations within a distance range of the delivery location. 
     
     
       19. The computer system of  claim 14 , wherein the one or more memories store additional computer-readable instructions that, upon execution by the one or more processors, further configure the computer system to:
 input an image to a second AI model, the second AI model trained for drop-off image quality evaluation based at least in part on the training images; 
 receive, from the second AI model and based at least in part on the image, an image quality evaluation; and 
 send, to the device, second data that comprises the image quality evaluation. 
 
     
     
       20. The computer system of  claim 19 , wherein the second AI model is further trained for image quality improvement based at least in part on an image attribute, the image attribute comprising at least one of: blurriness, brightness, angle relative to an item drop-off, relative location of the item drop-off in the image, or contextual background around the item drop-off, and wherein the one or more memories store further computer-readable instructions that, upon execution by the one or more processors, additionally configure the computer system to:
 receive, from the second AI model and based at least in part on the image, a recommendation about an image quality improvement, wherein the second data comprises the recommendation.